How we implement AI
Our approach is to always be testing the limits of AI but only implementing what is tried and tested. This way we maximise the gains whilst minimising the risks.
The 4 Levels of AI
Level 1:
Prerequisites for AI
The first layer isn’t AI at all, but a necessary prerequisite: optimising existing data flows. This involves structuring, cleaning, and connecting your systems so information moves reliably across the business. Without this foundation, introducing AI often leads to inconsistent or low-quality results.
Step 2:
Human Augmentation
The second layer is human augmentation. Here, AI acts as a tool that enhances human productivity while remaining fully under human control. It can draft emails, search and summarise information, analyse data, and surface useful insights - allowing people to work faster and more effectively without removing decision-making from their hands. We move beyond ChatGpt!
Step 3:
Semi-Autonomous Agents
​The third layer is semi-autonomous AI. At this stage, AI systems are given a limited set of predefined actions and use context to decide which to take. This enables automation of workflows and routine decisions, while still keeping the system within clear, controlled boundaries.
Step 4:
Autonomous Agents
The fourth layer is autonomous agents. You may have heard of OpenClaw - these are more experimental systems capable of developing their own strategies and ways of solving problems, rather than strictly following predefined paths. While powerful, they are not yet reliable or predictable enough for most commercial use, and so require careful consideration before deployment.